Spaces:
Sleeping
Sleeping
|
Download README.md from BonusLockSMith/csv-insights: direct link, hf CLI and curl.
- Browser
- Download file 2.22 kB
-
https://huggingface.co/spaces/BonusLockSMith/csv-insights/resolve/main/README.md
- Command line
-
hf download hf://spaces/BonusLockSMith/csv-insights/README.md
-
curl -L -o README.md https://huggingface.co/spaces/BonusLockSMith/csv-insights/resolve/main/README.md
2.22 kB
| title: CSV to Insights Analyst | |
| emoji: π | |
| colorFrom: gray | |
| colorTo: red | |
| sdk: docker | |
| app_port: 7860 | |
| pinned: false | |
| # CSV β Insights Analyst π | |
| Upload a CSV β get **auto-generated charts + a plain-English briefing you can trust**. | |
| > Project **#12** of the *"30 AI Projects in 15 Days"* build-in-public challenge. | |
| > Focus: **code execution + a data agent** (vs. #11's long-running service). | |
| ## Honest by construction | |
| The AI **never invents a number.** A deterministic pandas pass computes *every* statistic and chart; | |
| the model only turns that computed profile into prose. Units the model tries to guess (mm, Β°C, β¦) are | |
| stripped β they aren't in your file. Same grounded discipline as the #10 RAG app. | |
| ## What it does | |
| - **Profile** β shape, dtypes, missing %, numeric summaries, top categories, correlations, outliers, | |
| date-range detection, and plain-language data-quality flags. | |
| - **Charts** β trend-over-time, correlation heatmap, distributions, and category bars (GritAI design system). | |
| - **Findings** β a short, plain-English briefing grounded only in the computed numbers, with follow-up | |
| questions the data could answer next. | |
| ## How it works | |
| - `profiler.py` β computes the full structured profile (pandas). The source of every number. | |
| - `charts.py` β matplotlib charts from the profile, returned as inline base64 PNGs. | |
| - `narrator.py` β turns the profile into prose. Local Ollama by default (free/private); Claude optional. | |
| A deterministic post-pass strips any units the model added. | |
| - `app.py` β Flask UI: upload or try the sample, 12 MB cap, data handled in-memory (never persisted). | |
| ## Two tiers | |
| - **This hosted demo** narrates via Claude-haiku (charts + stats are 100% local pandas). | |
| - **GritAI Studio** (paid) runs the whole thing β narrative included β on your own local GPU fleet: | |
| free per-analysis, fully private, nothing leaves your network. | |
| ## Run locally | |
| ```bash | |
| pip install -r requirements.txt | |
| python app.py # http://127.0.0.1:7860 | |
| # Backend: set INSIGHTS_LLM_BACKEND=ollama (default, local) or =claude (set ANTHROPIC_API_KEY) | |
| ``` | |
| --- | |
| Built by **Robert Lucyk** Β· [GritAI Solutions](https://gritai.solutions) Β· part of the 30-in-15 challenge. | |